A third perspective on the effects of general health checks may provide a less biased estimate. Author response

Bruun Larsen et al. have performed a CACE analysis, which compares mortality among actual participants of the intervention group with a hypothesized group of participants in the control group in the Inter99 study [[1]Larsen L.B. Thilsing T. Sondergaard J. Bjerregaard A.-L. A third perspective on the effects of general health checks may provide a less biased estimate (letter commenting J Clin Epidemiol 2016;71:120–2).J Clin Epidemiol. 2018; (https://doi.org/10.1016/j.jclinepi.2018.05.018)Abstract Full Text Full Text PDF PubMed Scopus (3) Google Scholar]. CACE analyses could be a valuable supplement to existing analyses, if we did not have the results from an intention-to-treat (ITT) analysis, but we have these results making CACE analyses unnecessary. The authors suggest on the basis of their CACE analyses that there is a beneficial effect of the Inter99 study on total mortality among those who participated in the health check and lifestyle intervention. This could be so, but still an overall effect is absent, and we also will question their analyses: In the Inter99 study, there was an oversampling of middle-aged persons in the intervention group, and proportion of persons aged 55 or 60 years were considerably higher in the control group (31%) compared to participants in the intervention group (24%). Other confounders may as well be unevenly distributed between participants in the intervention and control group [[2]Bender A.M. Jørgensen T. Pisinger C. Is self-selection the main driver of positive interpretations of general health checks? The Inter99 randomized trial.Prev Med. 2015; 81: 42-48Crossref PubMed Scopus (26) Google Scholar]. Therefore, the analyses performed by the authors are heavily biased—close to useless. We disagree with the authors that results from CACE analyses are more reliable than those from ITT analyses. ITT analyses in Inter99 study cannot be subject to bias when, as randomization was blinded and there was no missing outcome measures. We find it misleading to use the term noncompliance bias in relation to ITT analyses, as this is a premise of ITT analyses: assessing effects of randomized studies disregard of participation, compliance, and maintenance of treatment. ITT analysis therefore provides information about the actual effects of a health policy. At the population level, results from CACE analyses reflect results that would only be seen in a motivated population resembling participants of the intervention group (significantly older than nonparticipants, a larger proportion was of Danish/Western origin, had a better socioeconomic profile and were healthier), but it is a question whether we could obtain better results if there was a 100% participation among invited persons. This was actually questioned in our analyses of those areas with high participation rates [[3]Bender A.M. Jørgensen T. Pisinger C. Do high participation rates improve effects of population-based general health checks?.Prev Med. 2017; 100: 269-274Crossref PubMed Scopus (6) Google Scholar] showing a deleterious effect among women. The question is not only whether it is possible to motivate nonparticipants to join such studies but also whether it will be of any benefit. Studies of individuals have in accordance with the CACE analyses results showed that changes in lifestyle can lower the risk of ischemic heart disease and mortality on an individual level. But what is effective at an individual level under ideal circumstances does not work when scaled up to the population level [[4]Krogsbøll L.T. Jørgensen K.J. Grønhøj Larsen C. Gøtzsche P.C. General health checks in adults for reducing morbidity and mortality from disease.Cochrane Database Syst Rev. 2012; : CD009009PubMed Google Scholar]. A third perspective on the effects of general health checks may provide a less biased estimate (letter commenting J Clin Epidemiol 2016;71:120–2)Journal of Clinical EpidemiologyVol. 102PreviewWe welcome the findings from Bender et al. on the differences in effect between the two analytical perspectives on general health checks detailed in the Inter99 study [1]. They present estimates of effects on total mortality and cardiovascular mortality based on the intention-to-treat (ITT) principle and on a participant-only analysis that compare participants to the intervention with participants in a subgroup of the control group that merely received a questionnaire on lifestyle. The analyses show a hazard ratio (HR) of 0.99 of death from the ITT analysis and an HR of 0.63 from the participant-only analysis. Full-Text PDF Response to letter to editor “Only ITT analysis provides information about the actual effects of a health policy”: Assessment of health policy effects of health checks requires a broader perspective than the ITTJournal of Clinical EpidemiologyVol. 107PreviewWe thank Bender et al for their reply and will briefly provide some comments. With its firm focus on internal validity, a randomized controlled trial (RCT) is first and foremost used to estimate intervention effects. Why else use such a firm design? In our reply to the original letter, we merely argued that under the assumption of noncompliance bias, that is first, incomplete participation, and second, differences between actual participants and nonparticipants, then a complier average causal effect (CACE) analysis will provide a less-biased estimate of the intervention effects (not the population effects) than an intention to treat (ITT) analysis and a per-protocol analysis [1]. Full-Text PDF